The style point multiplier doesn’t exist

Elegant solutions feel amazing. Sometimes they’re necessary.

And racking up boring, “we all know this is a good idea” wins will get us further than we’d like to admit.

A smattering of boring wins

Getting logs into JSON format

You don’t often get to make something cheaper, more discoverable, and more usable all at once. Putting your logs in JSON gets you all that.

Using a smaller model

If you’re using an LLM, could it be a boosted tree? If you’re using a boosted tree, could it be a linear model? If you’re using a linear model, could it be a set of deterministic rules?

Smaller models are easier + cheaper to deploy, maintain, and explain. The temptation is to always push for better model performance while neglecting those trade-offs.

Writing stuff down

It’s never been easier to turn implicit knowledge into durable text. Use text to speech to yap into a Google Doc. Have an LLM interview you about that tricky procedure your team members need to learn. Go wild and spend more time writing out your plans by hand.

Having ideas accessible to others was always better than having it live in our heads. Now we have tools that can leverage that text more directly than ever.

A boring win to you might be a revolution elsewhere

All the wins listed above are common knowledge in one field or another. But there’s a chance one of them could blow your team’s mind. People can be great at what they do and never hear common ideas in adjacent fields. Docker doesn’t 100% solve analysis reproducibility, and it comes a lot closer than I would have guessed while siloed in clinical psychology.

Let’s go out and put some boring wins on the scoreboard. Our work is hard enough without leaving opportunities on the table.

A data thing I liked

A thread of data engineering resources

A not-data thing I liked

Every disease is a policy failure